About Project
Simulation has been shown to be an effective tool for decision making and it has been used successfully in many different domains such as manufacturing, aerospace, maritime as well as defence. However, a typical simulation project life cycle is a long process and involves manual processes in model development, verification and validation, experiments design, deployment and execution, and final result analysis. The long simulation life-cycle is not agile and flexible enough to response promptly to the highly complex and dynamically changing environments that are typical in the context of defence operations.
In the defence domain, simulation studies are often focused on generating tactical and operational plans that will improve certain aspects of a system or operational strategies, with respect of a fixed set of assumptions (e.g. strength and tactics of enemies). The simulation results obtained from these simulation studies are often not robust and will not be valid when any of these assumptions are no longer true. To improve the robustness, flexibility and speed of simulation studies, this project will investigate the use of evolutionary techniques for simulation modelling and analysis.
Figure 1: Evolutionary Computing Platform for Modelling, Simulation and Analysis
As shown in Figure 1, the focus of this project is to develop an evolutionary computing-based platform for modelling, simulation and analysis. This involves the investigation and development of evolutionary computing based techniques that will be used in Automated Red Teaming, Automated Co-Evolution and Evolvable Simulation to enable more flexible and robust simulation modelling and analysis to be carried out.
Red teaming is a technique that is commonly used to uncover system vulnerabilities. Weaknesses in the system can then be fixed so that the robustness of the entire system can be improved. Preliminary studies have been carried out in the automation of this process with the new technique of Automated Red teaming (ART) via simulation-based parameters tuning using evolutionary computing approach. The selection of the evolutionary algorithm used in ART is important and this aspect has not been addressed by the research community. One of the focuses of this project is to explore new evolutionary algorithms such as Multi-Objective Particle Swarm optimization (MOPSO) and Bee Colony Optimization (BCO) that can be used in ART and compare them in terms of their speed, flexibility and accuracy.
While ART provides the mechanism for exposing the weaknesses of the Blue team, it does not provide solutions for fixing the weaknesses or making the Blue team more robust against an adaptive Red team. With ART, these solutions still have to be found through manual means by analysing those scenarios that the Red team outperform the Blue. To automate this process, the technique of Automated Co-Evolution (ACE) can be used to carry out simulation-based parameters tuning for the plans and strategies used by both the Red and Blue teams. This process is repeated until a set of robust strategies for the Blue team can be found. The ACE technique, which combines ART for both the Red and Blue teams, is currently being investigated by the ORL lab in DSO.
As ACE is a relatively new technique, there are many research issues that need to be resolved. This research will focus on two important research issues in ACE. The first research issue involves studying how multi-objective plans can be generated using ACE (existing work on ACE focuses only on single objective plan). The second research issue involves studying how the ACE process can be accelerated given the large number of simulation experiments to run.
To enable simulation-based parameters tuning using ART and ACE, a set of configurable parameters have to be implemented and enabled in the simulation model used in the study. This set of configurable parameters thus sets the limit and flexibility on the solution space for both the ART and ACE. For example, if the simulation model is implemented such that the Red team only attacks the Blue using routes that consist of waypoints from a pre-defined set, it is not possible for ART and ACE to generate solutions which have routes that use other waypoints not in the set and to evaluate their impact on the Blue team.
Evolvable simulation (EvoSim) is a new approach for simulation modelling, where simulation models can be generated and modified on-demand during the optimization and analysis process in ART and ACE. This idea was first mentioned by Upton but till date no further work has been reported in this area. The focus of this part of the project is to investigate the mechanisms needed for the realization of EvoSim, and also to study how EvoSim can be used to dynamically modify the structure of a simulation model and re-generate the simulation model during the ART/ACE process. This will allow more powerful and robust ART and ACE to be carried out.